Negation/Affirmation Errors

AbstractContextual word embeddings extracted from pre-trained models have become the basis for many downstream NLP tasks, including machine translation automatic evaluations. Metrics that leverage embeddings claim better capture of synonyms …...